CounterPersona: Defending against Unauthorized Persona Skill Distillation in Append-Only Settings

CounterPersona proposes a defense method for append-only settings to prevent unauthorized persona skill distillation, in order to protect personal privacy and labor autonomy. It constructs targeted counter-persona evidence, packages compatible behavioral states into compact implementation units, and reinforces them with rationale-guided consistency rewriting. Experiments show that the method is effective on lexical, semantic, and LLM-based metrics, and remains robust across different distillers.